saturn·

← all readings

processed lexibank references

source /home/coolhand/servers/diachronica/etymology_atlas/processed/lexibank_references.json 11,359 rows 9 columns profiled 2026-05-01 raw JSON static .html .ipynb Report Notebook

Reading

dataset summary · high confidence anthropic:claude-opus-4-7

This dataset is a bibliographic reference list with 11,359 rows and 9 columns (key, author, citation, title, year, plus mostly-empty editor/publisher/journal/url fields). The most informative columns are author, citation, title, and year — the rest are either unique IDs or near-empty categoricals. Note that author has a 66% duplicate rate and 1,277 empty values, while citation and title both show heavy duplication (54% and 50%) driven by a handful of large source collections like Koelle's Polyglotta africana and the Africa Museum and Austronesian web archives. The year column spans 271 distinct values with reasonable spread (entropy ratio 0.75), though about 11% of rows have no year and another 574 are marked 'n.d.'. Author and title are also multilingual, with English dominant but meaningful German, French, Spanish, and Chinese subsets.

citing: row_count · column_count · columns.author.stats.duplicate_rate · columns.author.stats.n_empty · columns.author.language_counts · columns.author.top_values · columns.citation.stats.duplicate_rate · columns.citation.top_values · columns.title.stats.duplicate_rate · columns.title.top_values · columns.title.language_counts · columns.year.n_unique · columns.year.stats.entropy_ratio · columns.year.top_values · columns.editor.stats.top_rate · columns.publisher.stats.top_rate · columns.url.stats.top_rate

year · Distribution of publication years; watch for the large empty/'n.d.' share and peaks around 1971, 1979, and 1992.
Show data table
valuecountshare
130011.4%
n.d.5745.1%
19923383.0%
20072822.5%
19712712.4%
19792542.2%
19802252.0%
20052211.9%
20152171.9%
19862081.8%
19972081.8%
20062041.8%
20092021.8%
20111961.7%
19751951.7%
1963 [1854]1931.7%
19811881.7%
20161881.7%
20041851.6%
20001851.6%
author · Top authors are dominated by a few prolific contributors (Koelle, Tryon, Blench), reflecting the 66% duplicate rate.
Show data table
charscount
0 – 112178
11 – 236163
23 – 341293
34 – 451034
45 – 57188
57 – 68206
68 – 79126
79 – 9130
91 – 10275
102 – 11331
113 – 1257
125 – 1364
136 – 1472
147 – 1595
159 – 1702
170 – 1812
181 – 1931
193 – 2040
204 – 2151
215 – 2260
226 – 2380
238 – 2490
249 – 2600
260 – 2720
272 – 2830
283 – 2943
294 – 3060
306 – 3176
317 – 3281
328 – 3400
340 – 3510
351 – 3620
362 – 3740
374 – 3850
385 – 3960
396 – 4080
408 – 4190
419 – 4300
430 – 4420
442 – 4531
citation · Citation string lengths cluster around 70 characters — useful for spotting truncated or anomalously short entries.
Show data table
charscount
8 – 117
11 – 1522
15 – 1820
18 – 218
21 – 2554
25 – 28128
28 – 3199
31 – 3599
35 – 38133
38 – 42125
42 – 4561
45 – 48269
48 – 5286
52 – 55119
55 – 5895
58 – 6271
62 – 65105
65 – 68312
68 – 725296
72 – 753139
75 – 78854
78 – 82168
82 – 8564
85 – 884
88 – 926
92 – 953
95 – 986
98 – 1020
102 – 1052
105 – 1080
108 – 1120
112 – 1151
115 – 1191
119 – 1220
122 – 1250
125 – 1291
129 – 1320
132 – 1350
135 – 1390
139 – 1421
title · Title lengths are highly skewed (median 104, max 1562); long outliers often contain embedded URLs.
Show data table
charscount
0 – 391194
39 – 782283
78 – 1173425
117 – 1561892
156 – 1951094
195 – 234646
234 – 273344
273 – 312178
312 – 35181
351 – 39056
390 – 43039
430 – 46928
469 – 50823
508 – 54724
547 – 58613
586 – 62514
625 – 6645
664 – 7033
703 – 7423
742 – 7812
781 – 8206
820 – 8591
859 – 8980
898 – 9372
937 – 9761
976 – 10150
1015 – 10540
1054 – 10930
1093 – 11320
1132 – 11720
1172 – 12110
1211 – 12500
1250 – 12890
1289 – 13280
1328 – 13670
1367 – 14060
1406 – 14451
1445 – 14840
1484 – 15230
1523 – 15621
title · Language mix of titles shows English dominant but with sizeable German, French, Spanish, and Chinese minorities.
Show data table
charscount
0 – 391194
39 – 782283
78 – 1173425
117 – 1561892
156 – 1951094
195 – 234646
234 – 273344
273 – 312178
312 – 35181
351 – 39056
390 – 43039
430 – 46928
469 – 50823
508 – 54724
547 – 58613
586 – 62514
625 – 6645
664 – 7033
703 – 7423
742 – 7812
781 – 8206
820 – 8591
859 – 8980
898 – 9372
937 – 9761
976 – 10150
1015 – 10540
1054 – 10930
1093 – 11320
1132 – 11720
1172 – 12110
1211 – 12500
1250 – 12890
1289 – 13280
1328 – 13670
1367 – 14060
1406 – 14451
1445 – 14840
1484 – 15230
1523 – 15621

Schema

9 columns
Alerts
key text 0.0% 11,359
near_unique one_word allcaps short_text
author text 0.0% 3,830
multilingual duplicates
year categorical 0.0% 271
title text 0.0% 5,663
multilingual duplicates
journal categorical 0.0% 4
imbalance
publisher categorical 0.0% 15
long_tail imbalance
editor categorical 0.0% 4
long_tail imbalance
url categorical 0.0% 1
imbalance
citation text 0.0% 5,178
multilingual duplicates

key

text identifier near_unique one_word allcaps short_text
This column is almost certainly a primary identifier: every one of the 11,359 rows holds a unique, single-token value (n_unique=11359, one_word_rate=1.0, duplicate_rate=0.0). Values are short (len_mean 4.07, len_max 5) and 99.1% are uppercase, consistent with short alphanumeric codes rather than natural text. The top_words sample shows purely numeric tokens (47, 49, 50, ...), so the 'allcaps' signal may be a side effect of digit-only strings rather than true letters. Treatment: Use as a row key or left-join key; drop from modelling features. high · anthropic:claude-opus-4-7
n
11,359
nulls
0 (0.0%)
unique
11,359
len_min
1
len_max
5
len_mean
4.074
len_median
4
len_p95
5
word_mean
1
word_median
1
n_empty
0
n_duplicates
0
duplicate_rate
0
vocab_size
11,359
readability_flesch_mean
121.2
emoji_rate
0
url_rate
0
one_word_rate
1
allcaps_rate
0.9913
boilerplate_rate
0

author

text metadata multilingual duplicates
This is an author/contributor name field, mostly formatted 'Surname, Given Name' with frequent multi-author strings joined by 'and' (4049 occurrences). Duplication is severe: 66.3% of rows repeat, with prolific contributors like 'Koelle, Sigismund Wilhelm' (193) and 'Tryon, Darrell T.' (191) dominating, and 1277 rows are empty strings despite a 0.0 null rate. Names span 30 detected languages — predominantly English (3125) and German (384), but also 113 rows of CJK script — so naive string matching will fragment identities. Treatment: Normalize to a canonical name form and split multi-author strings on ' and ' before any join or grouping. high · anthropic:claude-opus-4-7
n
11,359
nulls
0 (0.0%)
unique
3,830
len_min
0
len_max
453
len_mean
20.43
len_median
17
len_p95
50
word_mean
3.382
word_median
3
n_empty
1,277
n_duplicates
7,529
duplicate_rate
0.6628
vocab_size
6,656
readability_flesch_mean
53.21
emoji_rate
0
url_rate
0
one_word_rate
0.1836
allcaps_rate
0.0589
boilerplate_rate
0

year

categorical feature
This is a year field stored as strings rather than integers, with 271 distinct values across 11,359 rows. The most common entry is an empty string (1,300 rows, 11.4%) followed by the literal "n.d." (574 rows), so roughly 16.5% of records carry no usable year despite a 0.0 null rate. Actual years span at least 1971 to 2015 in the top values, with 1992 the most frequent real year at 338 occurrences. Treatment: Coerce to integer, mapping empty strings and "n.d." to missing before any temporal analysis. high · anthropic:claude-opus-4-7
n
11,359
nulls
0 (0.0%)
unique
271
top_value
top_rate
0.1144
cardinality
271
entropy
6.1
entropy_ratio
0.7548

title

text metadata multilingual duplicates
This column holds bibliographic citation strings — book/article titles, URLs, and access notes for linguistic sources, dominated by English (3620) but mixing 24 other languages including French (382), Chinese (300), German (224), and Spanish (207). Half the values are duplicates (duplicate_rate 0.50, 5696 repeats across 11359 rows), with a single africamuseum.be URL appearing 355 times and 17.4% of entries containing URLs. Length varies wildly (median 104 chars, max 1562) and top words ('of','the','languages','linguistics','(accessed') confirm these are reference citations rather than free-form titles. Treatment: Normalize and deduplicate citations into a source-reference lookup table rather than treating as a modelling feature. high · anthropic:claude-opus-4-7
n
11,359
nulls
0 (0.0%)
unique
5,663
len_min
0
len_max
1,562
len_mean
120.6
len_median
104
len_p95
261
word_mean
14.66
word_median
12
n_empty
7
n_duplicates
5,696
duplicate_rate
0.5015
vocab_size
21,846
readability_flesch_mean
8.003
emoji_rate
0
url_rate
0.1741
one_word_rate
0.07008
allcaps_rate
0.06154
boilerplate_rate
0

journal

categorical metadata imbalance
This appears to be a journal-name field for bibliographic records, but it is effectively empty: 11,355 of 11,359 rows (top_rate 0.9996) hold an empty string, leaving only 4 actual journal names across 3 distinct German linguistics titles. Entropy is 0.005 (entropy_ratio 0.0025), so the column carries virtually no information despite a 0.0 null rate — the blanks are stored as empty strings rather than nulls. Treatment: Drop; near-constant empty string with only 4 populated rows. high · anthropic:claude-opus-4-7
n
11,359
nulls
0 (0.0%)
unique
4
top_value
top_rate
0.9996
cardinality
4
entropy
0.005076
entropy_ratio
0.002538

publisher

categorical metadata long_tail imbalance
Publisher name field, but it is effectively empty: 11,344 of 11,359 rows (top_rate 0.9987) carry an empty string, leaving only 15 distinct values and an entropy_ratio of 0.005. The handful of populated entries (Brill with 2, then Winter, Reichert, Rodopi, Harrassowitz and others with 1 each) hint at academic/humanities publishers but are too sparse to be useful. Note that null_rate is 0.0 because the blanks are stored as empty strings rather than true nulls. Treatment: Drop; the column is ~99.9% empty strings and carries almost no signal. high · anthropic:claude-opus-4-7
n
11,359
nulls
0 (0.0%)
unique
15
top_value
top_rate
0.9987
cardinality
15
entropy
0.01952
entropy_ratio
0.004996

editor

categorical metadata long_tail imbalance
This appears to be an editor name field for bibliographic records, but it is effectively empty: 11356 of 11359 rows (top_rate 0.9997) hold the empty string, with only three distinct named editors each appearing once. Entropy is essentially zero (0.0039) and cardinality is just 4, so the column carries almost no information despite a 0.0 null rate (blanks are encoded as ''). The long_tail and imbalance alerts simply reflect that three singleton names sit beside one dominant blank. Treatment: Drop; near-constant blank with only three populated rows. high · anthropic:claude-opus-4-7
n
11,359
nulls
0 (0.0%)
unique
4
top_value
top_rate
0.9997
cardinality
4
entropy
0.003939
entropy_ratio
0.001969

url

categorical metadata imbalance
This column is labelled 'url' but contains a single value—an empty string—across all 11,359 rows. Cardinality is 1, entropy is 0, and the top_rate is 1.0, so it carries no information whatsoever. Likely a placeholder field that was never populated during ingestion. Treatment: Drop; the column is constant and has zero predictive value. high · anthropic:claude-opus-4-7
n
11,359
nulls
0 (0.0%)
unique
1
top_value
top_rate
1
cardinality
1
entropy
0
entropy_ratio
0

citation

text metadata multilingual duplicates
Bibliographic citation strings for linguistic sources, mostly English (3590) with substantial French (401), Chinese (335), and German (249) entries. Heavy duplication is the headline: 6181 duplicates (54.4%) across only 5178 unique values, with the top value 'Anon. n.d.. http://www.africamuseum.be/...' repeating 462 times. URLs appear in 13.3% of rows and Flesch readability is low (29.3), consistent with reference-style text rather than prose. Treatment: Normalize and deduplicate to a citation lookup table, then reference by key. high · anthropic:claude-opus-4-7
n
11,359
nulls
0 (0.0%)
unique
5,178
len_min
8
len_max
142
len_mean
67.94
len_median
71
len_p95
77
word_mean
8.827
word_median
10
n_empty
0
n_duplicates
6,181
duplicate_rate
0.5442
vocab_size
15,535
readability_flesch_mean
29.26
emoji_rate
0
url_rate
0.1334
one_word_rate
0
allcaps_rate
0.06004
boilerplate_rate
0